The Attribution Signal Loop Meta Uses for Optimisation Explained

Quick Answer

Meta's optimisation engine relies on a closed feedback loop: impressions → conversions → labelled training data → updated eAR predictions → better impressions. Every conversion event you report becomes a labelled training example for the model. The cleaner and faster your signal, the tighter the loop, the better the optimisation. CAPI matters because it closes the loop server-side where browser tracking has been broken since iOS 14.5.

The Mechanism Explained

The loop has four stages:

1. Impression served — Meta logs (ad_id, user_id, timestamp, context features)
2. User takes action — pixel fires (browser) or CAPI fires (server) with the conversion event
3. Attribution match — Meta links the conversion back to the impression using the click ID, user ID, or fbclid
4. Model retraining — the matched (impression, conversion) tuple becomes a positive training example for eAR

Critically, impressions without matched conversions become negative training examples. This means you don't just need to log conversions — you need to log them in a way Meta can match. Unmatched conversions are double-bad: they don't reinforce your ad, and the impression that caused them gets labelled as a negative.

The signal quality factors:

• Match rate — what % of your conversions Meta successfully attributes to an impression. Above 70% is good, above 85% is excellent.
• Latency — time from conversion to logged event. Under 2 hours is healthy.
• Deduplication — pixel + CAPI events for the same conversion should be deduped via event_id or external_id. Mismatch = inflated event count, polluted training.
• Coverage — what % of your conversion events make it to Meta at all. iOS users without CAPI typically lose 30-50% of conversions.

In 2026, Conversions API Gateway has become the standard CAPI implementation, with most ecom platforms shipping it natively. Without CAPI, your optimisation runs on degraded signal, and your ad set's effective eAR is lower than it should be.

Practical Implication

If you don't have CAPI configured properly, fix that before doing anything else with bidding or creative. Every improvement downstream is multiplied by signal quality. A 50% match rate cuts your effective optimisation in half regardless of how well you set everything else.

Real Numbers

• Match rate without CAPI on iOS-heavy traffic: 40-55%
• Match rate with proper CAPI + dedup: 80-92%
• CPA difference between match-rate >80% and <60%: typically 25-40% in favour of high match rate

FAQs

Q: Does CAPI replace pixel?
No — Meta recommends both, deduped via event_id.

Q: How fast should CAPI events fire?
Within minutes of the action — under 1 hour ideal.

Q: Can offline conversions enter the loop?
Yes — via Offline Events upload or CAPI for offline.

Q: Does Meta penalise low match rate?
Indirectly — your eAR suffers because the model has worse training data on you.

Q: Will using CAPI improve learning phase exit?
Yes — match rate is one of the strongest predictors of fast exit.

Pix-Vu

Once your signal loop is clean, creative is the next biggest lever. Pix-Vu helps you produce the creative variants that get labelled as positives in the feedback loop — at https://pix-vu.com.